This is the development record, published as it was written. It is amended by revision, including where the work went wrong. Nothing here has been rewritten for the web.

Record header

created_at: 2026-09-19T19:38:00-06:00
format: perspicuity-work/1
id: ae-2026-09-19-cycle-one-delivery-scope
next_check: 2026-10-03
record_status: open
revision: 1
skill_version: 0.4.0
updated: 2026-09-19
updated_at: 2026-09-19T19:38:00-06:00
work_status: active

Choose what cycle one delivers

Current position

Parent: RECORD.md, revision 6.

Principal: David. Decider: Rook for the cycle-one delivery scope and the reversible implementation choices inside it, under the delegation in RECORD.md; David for the price, the outreach, the site's domain, the go-live word and anything published about a supplier. Work owner: Rook. Decision: selected — A2, the closed journey: today's product plus the minimum that lets a stranger arrive, submit, read, take the files away and share, and nothing else. selected_at 2026-09-19T19:42:00-06:00, by Rook, against revision 1 of this record, which was registered and committed at 72c1810 before the comparison was written. Authorisation: David's instruction of 2026-09-19 to start the next phase from the core delivery decisions, inside the delegation in RECORD.md. Work scope: what cycle one builds and runs, what it deliberately does not build, and what "live" means for the registered market test. It does not own the price, the outreach message, the domain choice or the go-live word. Work: registered at 19:38 with the frame, the objectives, the conditions and the alternatives; selected at 19:42 after three real supplier audits were run as U1. U1's evidence is in Act. No product code has been written in this increment. Outcome: unknown, and nothing is claimed. No supplier has submitted a domain; the site is not hosted; no supplier has paid. Next: Rook — build U3–U5, holding the corpus in step, and report to David at go-live. David — choose the domain and give the go-live word; both remain his, and the domain is the only thing on the critical path that this record cannot move. Dependency: the site's domain (David, parent uncertainty U3) gates U5 but not U3 or U4. Nothing else blocks. Review due: 2026-10-03, the scope cap registered in the parent's plan; the parent's test window opens at publication.

Registration note. This file was committed at 72c1810 with the frame, the objectives, the conditions and the alternatives, and *without* a recommendation or a selection. The comparison and the selection were written afterwards, at 19:42, from evidence gathered in between. The two commits are the evidence that the basis was saved before the evaluation: git show 72c1810 is the registered basis, and the diff to this revision is what the evidence changed. This is the practice docs/RECORDS.md asks for, and unlike the previous increment it was followed.

Frame and Decide

David accepted the process increment on 2026-09-19 and instructed Rook to start the next phase, beginning with the core decisions about what the project will actually deliver. The inherited material is the parent's selection and scope decisions S1–S7, which are not repeated here.

The question this record answers:

What is the smallest delivery that makes a real small supplier able to get a useful report, take the files away, and share it — without adding a dependency, an account, a model or a claim we cannot support?

The framing that matters, and why it is narrower than it first appears. The product's parts are not equally proven. The check engine exists and is deterministic. The report exists and renders. The journey — a stranger arriving, entering a domain, waiting, reading, acting, sharing — has never been walked end to end by anyone outside this project, and walking it is what produces the ten real audits the market test counts. So cycle one is not "finish the product". It is close the journey and get out of the way of the test.

Fundamental objectives

Material conditions

Alternatives and consequences

The four courses registered before evaluation, now compared against the objectives and the conditions.

The decisive tradeoff. A1 and A2 both make the test runnable; A2 costs more and is the only one that tests O2 at all. The preference it rests on is that a walkable journey is worth more than a smaller build, because the journey is the thing being sold and the thing the market test measures — ten audits by people who are not us. A1 would produce ten audits too, but as a consultancy, and the finding would be about David's outreach rather than about whether a supplier can use the thing unaided.

A3 is rejected for the window, not for the idea: it is the revenue hypothesis, and building it before a single supplier has seen a report would be building the paid answer to an unasked question. A4 is deferred rather than rejected — an agent-facing interface is the product's own logic, but building one with no consumer is the same mistake in a different shape.

What would warrant reconsideration. If the domain cannot be chosen within a few days, A1 becomes the better course: it is the only one that can start the test without the domain, and the window is the binding constraint. If, at the first real audits, suppliers consistently cannot find or use the report unaided, the answer is not a bigger build but a plainer report — recorded as a revision here rather than as scope.

Selection

selected_at 2026-09-19T19:42:00-06:00, by Rook, against revision 1 committed at 72c1810. A2.

Three things are chosen with it, and they are the consequential part:

1. Publishing is a server action, not a link copy. A supplier who chooses to publish gets a stable, human-readable, indexed URL for a *copy* of the report, taken at a recorded instant; the original report stays at its unguessable id and stays out of the index. This changes what we publish, so it is the decision in this record that most deserves its own paragraph. It also means the "private by default" stance survives with its meaning intact: nothing becomes public without an explicit action by the supplier, and the public copy is dated so a reader can tell how old it is.
2. The report's share state is visible on the report. A published report says so and says when. A supplier who has not published sees that it has not — which is both honest and the prompt that makes the action discoverable.
3. The paid tier is described, not built, and described as unavailable. One paragraph on the landing page, no price, no date, no mechanism. Building a "notify me" field would be collecting contact details from people we have promised not to track, and agreeing a price is David's. The consequence to accept: cycle one produces no revenue path at all, and the market test's revenue half can only be observed through enquiries that arrive by whatever means David opens.

Rejected within A2, deliberately: an on-site index of published reports (a distribution decision disguised as a feature, and one that makes us a directory of our customers' weaknesses — David's to decide, not this record's); a plain-language pass over the report before any supplier has read one (we would be guessing at the confusion); and any e-mail capture (a standing constraint).

Act

U1 — the audit evidence this selection rests on

Run 2026-09-19 with python3 -m eligibility audit <domain>, at commit 72c1810, against three real supplier sites. The three were chosen from public domains already named in this workspace, not from a market sample.

Determinism, measured rather than asserted. goodmeat.co was audited twice, two seconds apart. The two outputs are byte-identical apart from the timestamp line: diff reports no differences across 19 lines.

What this establishes. The engine runs end to end on real supplier-shaped sites, the same input gives the same output, and the checks find things that are genuinely absent rather than merely unusual. The quorn.co.uk result — a declared sitemap that 404s — is the kind of finding a supplier cannot see by looking at their own site, which is the product's reason to exist.

What this does not establish, and the honest limit of U1: two of the three are large brands with in-house web teams, not small suppliers. The third is a small producer. So this is evidence that the tool works, not evidence about the target market, and it does not upgrade O1 or the payer hypothesis in any way. The small-supplier sample remains unrun.

The plan

What is deliberately not in this increment, recorded so that the scope fence is auditable: the monitoring tier (A3), any agent-facing interface (A4), an on-site index of published reports, any e-mail capture, any payment path, any plain-language report rewrite before a real reader has seen one, and any change to the 17 checks. Each is either David's or waits on evidence.

Acceptance criteria for the increment, registered before the work

1. make ci exits 0 and make records is clean at every commit.
2. U3 and U4 ship with tests that fail if the privacy stance is broken: a published copy is dated, and an unpublished report is neither indexed nor listed.
3. The site passes its own rubric once hosted, or the failure is recorded with its reason.
4. Every choice in U3–U5 that meets the admission test is in a record *before* the code that depends on it.
5. The published worked example is our own domain, audited by this tool, with its weaknesses shown.

Units registered before their implementation

U3–U5 are registered here as results and acceptance criteria. Their *design* choices — the publish endpoint's shape, what the published copy is, what the first-run experience says — meet the admission test and will be registered before the code that depends on them, in this record or in a linked one. This section is the plan, not the design.

Review

Changes

Revision 1, 2026-09-19T19:38:00-06:00. Created with the frame, objectives, material conditions and four candidate courses of action, and no selection. Source: David's acceptance of the process increment and his instruction to start the next phase from the core delivery decisions. Reason: the choice of what cycle one delivers governs every unit that follows, so its basis had to be saved before the comparison and before any code. Affects: the cycle-one plan, the units derived from it, and the domain decision's place on the critical path. No code was written and nothing was published, spent or sent.

Revision 2, 2026-09-19T19:42:00-06:00. Adds the comparison, the selection of A2, the plan and the review criteria. Source: three real supplier audits run between the two revisions, and the registered basis at 72c1810 read against them. Reason: the frame asked what the smallest delivery is that makes the journey walkable; the audits establish that the engine is not the constraint, so the journey is, and A2 is the only course that closes it. Changes: Alternatives and consequences replaces its placeholder with a comparison against the objectives and conditions; Selection records A2 and the three choices taken with it, including the publish decision that changes what the project may publish; Act carries U1's evidence, the U1–U6 plan, the acceptance criteria and the explicit scope fence; the Review criteria are registered. Revision 1 is preserved at 72c1810. No objective or material condition in revision 1 is altered. Nothing is published, spent or sent; no product code was written.